A characterization of Arctic undersea noise

Arctic undersea noise is investigated and partially characterized by a probability model suggested by exploration of a sample of such noise. The large size of the data set makes this possible. The model is that of a mixed spectrum, with stationary Gaussian noise intermixed with random sinusoids, and occasional high intensity impulsive noise bursts, which are very short-lived. Due to the complex behavior of the random sinusoids, the noise cannot be considered as stationary in the time frame (10 min) investigated. The validity of the model is checked by testing model predictions against the actual data. In particular, this model makes some predictions about the type of spectrum observed, and these predictions are tested by a nonlinear filtering of the periodogram. A running median filter is used and is shown to asymptotically estimate the spectrum of the Gaussian noise. This filter seems to work fairly well on the data.